Three more Pussyfootin' A.I. episodes have gone out since the last catch-up post, so here they are in one place. Same rules as always: real things I actually did with AI, under a minute each, with the whole recipe given away free.
Side by side, they all land on the same idea. AI did the making every time. What made it work, or caught where it didn't, was something I already knew from twelve years of editing: how to write a shot list, how to watch continuity, and how to spot a sentence about me that isn't quite true.
My cat starred in a movie made from four phone photos
My cat just starred in a movie and slept through the whole shoot.
I took four photos on my phone: three of my cat and one of my desk. I dropped them into Creo as ingredients, wrote the prompt like a shot list (wide, close-up, insert, hero), and picked the look from its cinematography menu: an anamorphic lens and tungsten film stock. Seedance 2.0 built all four shots in one 12-second clip, for $6.72.
How it works: Creo puts several video, image, and audio models behind one interface. I used Seedance 2.0's Ingredients to Video mode, which takes up to four reference images and treats them as your cast and your set. You describe the sequence and the model builds it with your subject in it. The cinematography menu adds lens and film choices, the way you'd brief a camera crew.
Why a shot list works: I've been editing for twelve years, and it turns out I was writing prompts the whole time. A shot list already gives a model what it needs: the framing, one action per shot, and where the cuts go. Start with one paragraph that sets the subject, the location, and the light for the whole sequence, then write one line per shot. The model honored all four hard cuts, in order.
The recipe: Creo, Seedance 2.0, Ingredients to Video, 1080p, 12 seconds. Ingredients: four phone photos, three of the cat and one of the desk. Lens: Anamorphic. Film: Tungsten 500T. The prompt, exactly as I ran it:
Cinematic 12-second sequence, four shots with hard cuts. The same cat from the reference photos in every shot, sitting on the desk from the reference photo, at night. Warm desk lamp from the left, cool blue monitor glow as a rim light on the fur.
SHOT 1, WIDE: dark room, slow push in. The cat sits on the desk, silhouetted against the glowing monitor.
SHOT 2, CLOSE-UP: the cat's eyes catch the screen light. One slow blink.
SHOT 3, INSERT: a paw reaches in and taps one key on the keyboard.
SHOT 4, HERO: the cat turns and looks straight into the lens. Hold.
Swap in your own subject and set, keep the structure.
The honest notes: Creo was in early access when I made this, with no subscription, just a usage balance, so check current pricing before you plan around my number. Seedance 2.0 at 1080p was $0.56 a second, which is where $6.72 for 12 seconds comes from, and the pricing page said failed generations are refunded automatically. And to be clear, my cat is real. None of the footage is.
The continuity lesson from my AI short film
I made the characters. I made the scenes. I wrote the script. And AI still couldn't keep her sunglasses in the same place.
In September I made a short film with AI: "I Just Want to Get Home to My Cat," built shot by shot in Higgsfield's Cinema Studio with Kling 3.0 and Seedance 2.5. Some of it was great. Her face held in the dialogue shots, and the lines synced. But cut it together and it falls apart. Her café turns into three different places. Her scarf comes and goes.
Why it happens: every shot is generated on its own. The model doesn't remember the shot before it, so anything you don't hand it again, it reinvents: the location, the wardrobe, the props. Then came the part that surprised me. I wrote "sunglasses over her eyes" in the prompt, but my reference photo had them pushed up on her head, and the model followed the photo, not my words. The reference image beats the prompt text.
The fix that worked for me: pick each character's reference image for what she's wearing in that shot. One reference per wardrobe state: shades on, shades up, scarf, no scarf. Then check every shot against the one before it before you generate the next.
A continuity pass for AI shots: this is the check I run now, the same job a script supervisor does on set. Before you generate the next shot, line it up against the last one and ask:
- Wardrobe: is she wearing exactly what she wore a cut ago? Glasses on or up, scarf or no scarf.
- Hair: up or down, same as the last shot.
- Location: same café, same table, same window behind her?
- The reference: does the image you're handing the model match this moment, not just this character?
The honest note: the models did the hard parts well. Her face held and the dialogue synced. What broke was continuity, which is the first job they teach you in film school. I have a film degree, and it turns out that's still the director's job.
A cover letter I didn't write
Do you write a cover letter when you apply for jobs? I don't anymore.
I handed one real application to Claude: a Senior Content Creator role at a creative agency. It researched the company, rewrote my resume for that one job (9 changes, every one from my real experience), wrote a 244-word cover letter, and filled out all 15 fields of the form. It didn't get the final word. I read both documents, and I clicked Submit myself.
How it works: Claude Code drove my own Chrome browser through Anthropic's Claude in Chrome extension. It read the posting, researched the company from its own site and work, built the tailored resume and the letter as PDFs, and filled in the form. Before any of that, it got a short set of rules, and the rules are the whole trick. Without them, a model will happily make you sound more impressive than you are.
The rules I gave Claude before it touched my resume:
- Change only four things: the headline, the 2-3 line summary, which bullets come first, and the order of my skills.
- Never invent. Every line has to come from my real resume.
- Keep every job title exactly as it is.
- Cover letter in plain paragraphs, no bullets. Open with the company, not me. One specific, checkable detail per claim.
- I read both documents before anything gets uploaded, and I click Submit.
The recipe: paste this into Claude Code with Claude in Chrome connected, and fill in the brackets.
I'm applying for [job title] at [company]. The posting is [link] and my resume is [file]. First, research the company from its own website and work, and tell me what you found, with links. Then tailor my resume for this one job. Change only four things: the headline, the 2-3 line summary, which bullets come first, and the order of my skills. Never invent: every line has to come from my real resume. Keep every job title exactly as it is. List every change you made. Then write a cover letter in plain paragraphs, no bullets, that opens with the company, not me, with one specific, checkable detail per claim. Stop and wait for me to read both documents before you upload anything. You can fill in the form, but I accept any terms and I click Submit.
The honest notes: decide up front what the AI is allowed to type. I gave it my answers for the personal fields (rate, address, work authorization) in the chat, so it never had to guess one, and they're blurred in the video. Accepting the privacy statement and clicking Submit stayed mine. Then read every line against your real resume. The failure to watch for is a verb that quietly gets promoted, "supported" turning into "led." AI can write about my twelve years. Only I can make sure it's true.
Follow along
Pussyfootin' A.I. lives on LinkedIn: one real move per episode, with the whole recipe given away free. No newsletter, no signup. The episodes land here too, with more room than LinkedIn gives me.
If you want media or workflows like this built for your own product or brand, by someone who treats the finish as the point, reach out.